Issue #4: 🟢 Implement Air Quality Forecasting Model

[ZEZE1020]

Description

Implement ML-based air quality forecasting to predict conditions 24-48 hours ahead.

Tasks

Phase 1: Simple Forecasting

• Install Prophet or ARIMA library
• Implement time-series forecaster
• Use historical OpenAQ data
• Generate 24-hour forecast
• Add confidence intervals

Phase 2: Feature Engineering

• Add weather features
• Add temporal features (hour, day, season)
• Add spatial features
• Normalize features

Phase 3: Model Training

• Train Random Forest or XGBoost
• Use the last 30 days of data
• Validate with the last 7 days
• Save model to disk
• Compute accuracy metrics

Phase 4: API

• Create forecast endpoint
• Load trained model
• Generate predictions on demand
• Cache predictions (1 hour)
• Return confidence intervals

Files to Create

backend/src/ml_models/
├── forecaster.py
├── trainer.py
├── feature_engineer.py
└── model_loader.py

API Endpoints

GET /api/v1/forecast/location?lat={lat}&lon={lon}&hours={hours}
POST /api/v1/forecast/train
GET /api/v1/forecast/accuracy

Alternative (if time-constrained)

Use a simple persistence model or linear regression instead of Prophet/ARIMA.

Acceptance Criteria

• Working forecast model
• 24-hour predictions generated
• Confidence intervals included
• API endpoints are functional
• Predictions cached
• Model accuracy documented

---